Building a Strategy for Sovereign AI in 2026 and Beyond
THE SCENARIO A government commissions a national AI strategy. The consultants deliver a roadmap: invest in compute, train talent, fund research, attract companies, regulate responsibly. The roadmap is approved. The budget is allocated. Three years later, the compute is foreign, the talent works for foreign firms, the research is published in foreign journals, the companies […]
The Hidden Truth About AI Benevolence Narratives
THE SCENARIO A technology leader announces a new AI initiative for global good. The initiative will solve climate change, cure disease, and educate every child. The model will be open. The compute will be donated. The governance will be multi-stakeholder. The sovereignty analysts read the fine print. The model weights are released under a license […]
Why We Need to Talk About the AI Supply Chain
THE SCENARIO An AI model is deployed in a critical national system. The model was trained on a framework from Company A. The framework runs on a runtime from Company B. The runtime executes on hardware from Company C. The hardware is manufactured in Country X. The fabrication equipment comes from Country Y. The design […]
The Sovereign’s Guide to Cross-Border Data Flows
THE SCENARIO A data flow agreement is signed between two allied nations. The agreement promises free flow of data with protections. The protections are based on a mutual adequacy determination. The determination is made by regulators who meet annually. The data flows. The protections are tested when a third-country court issues a data access order. […]
Rethinking: This Is Not a Technical Question (3)
THE SCENARIO A regulatory sandbox for AI is launched with great fanfare. The sandbox tests technical safety, bias metrics, and robustness benchmarks. The participants are technology companies. The regulators are technology regulators. The civil society observers are invited to comment on the technical reports. The governance question — who decides what safety means for this […]
The AI Workforce Gap: Training for Sovereignty Not Just Skills
THE SCENARIO A national AI skills programme trains ten thousand engineers in prompt engineering, model fine-tuning, and API integration. The graduation ceremony is a political showcase. Six months later, the graduates are employed by foreign AI labs, building foreign models, on foreign infrastructure. The skills are domestic. The sovereignty application is not. What is the […]
The Quiet Crisis of AI Procurement in Public Services
THE SCENARIO A municipal government procures an AI system for social services eligibility determination. The procurement follows all transparency rules. The contract includes data protection clauses, audit rights, and termination provisions. Two years later, the vendor is acquired by a foreign conglomerate. The new owner migrates the platform to its global cloud. The audit rights […]
Beyond the Hype: The Strategic Autonomy of Compute
THE SCENARIO A nation announces a sovereign AI compute initiative. The strategy focuses on acquiring GPUs, building data centres, and subsidising local cloud providers. The hardware arrives. The data centres open. The cloud services launch. Then a sovereignty audit reveals: the GPU firmware is controlled by the manufacturer. The data centre management software is licensed […]
What Most People Get Wrong About AI Standards Bodies
THE SCENARIO An international standards body publish a new AI safety standard. The standard is adopted by regulators worldwide. Compliance becomes a market requirement. The standard was drafted by a working group dominated by three companies. The test methodologies favour their architectures. The evaluation metrics reflect their priorities. The standard is voluntary. The market makes […]
The Real Cost of Ignoring Open Weight Models
THE SCENARIO A government commits to a closed-source AI platform for its critical public services. The vendor offers favourable terms, dedicated support, and regulatory compliance guarantees. Five years later, the vendor doubles the licensing fees, deprecates the API version the government depends on, and announces the model will no longer be updated for the government’s […]